On the Gittins Index for Multiarmed Bandits

This paper considers the multiarmed bandit problem and presents a new proof of the optimality of the Gittins index policy. The proof is intuitive and does not require an interchange argument. The insight it affords is used to give a streamlined summary of previous research and to prove a new result: The optimal value function is a submodular set function of the available projects.